Attention-Guided Answer Distillation for Machine Reading Comprehension
This addresses the problem of slow and attack-prone ensemble models for researchers and practitioners in NLP, offering a more efficient and robust solution, though it is incremental as it builds on existing distillation techniques.
The paper tackles the inefficiency and vulnerability of ensemble models in machine reading comprehension by using knowledge distillation to transfer ensemble knowledge to a single model, resulting in a student model with only a 0.4% F1 drop on SQuAD, 12x faster inference, and improved performance on adversarial datasets and NarrativeQA.
Despite that current reading comprehension systems have achieved significant advancements, their promising performances are often obtained at the cost of making an ensemble of numerous models. Besides, existing approaches are also vulnerable to adversarial attacks. This paper tackles these problems by leveraging knowledge distillation, which aims to transfer knowledge from an ensemble model to a single model. We first demonstrate that vanilla knowledge distillation applied to answer span prediction is effective for reading comprehension systems. We then propose two novel approaches that not only penalize the prediction on confusing answers but also guide the training with alignment information distilled from the ensemble. Experiments show that our best student model has only a slight drop of 0.4% F1 on the SQuAD test set compared to the ensemble teacher, while running 12x faster during inference. It even outperforms the teacher on adversarial SQuAD datasets and NarrativeQA benchmark.